Should I vibe code
Create one-page and multi-page resumes from structured evidence with layout constraints
Your resume renders beautifully in Chrome. The parser reading it sees one column of interleaved nonsense.
?
Their verdict, the Premium price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-04.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
Everything in this build belongs to you: your employment history, your machine, a PDF you read before anybody else does. There is no account to breach, no stranger's data to lose and no third party depending on it staying up, which is why the numbers come out where they do. The failure mode is not security, it is that the document leaves your hands and the first thing to read it is usually a parser rather than a person. What Novoresume actually sells is refusal — it will not let you set eight-point type, will not let a two-column layout swallow a job title, will not let page three happen. Your own builder will let you do all three cheerfully, and it will look immaculate in Chrome while doing it. So build it, keep the content in a structured file you own, and add the one test the paid tools quietly run for you: extract the text back out of your own PDF and read it in the order a machine would.
What actually breaks
not "if". the specific failures.
- Page breaks, which move the instant you add a line, slicing a role in half across two pages
- Font embedding: the file looks right on your laptop and substitutes to something else on every machine where the font is not installed
- Two-column layouts, which a text extractor reads as a single column, interleaving your dates with somebody else's bullet points
- Headless-Chrome rendering that differs between the version on your machine and the one in the container you ran it in at 11pm
- Typographic characters — the curly apostrophe, the en dash, the non-breaking hyphen — arriving as boxes in a recruiter's viewer
- PDF metadata you never looked at, where the Author field still carries the name of whoever wrote the template you started from
- Hyphenation and widow control, absent by default, which is most of the visual difference between a homemade CV and a bought one
Is that you?
the verdict is a default, not a law
- It runs locally and the only CV in it is yours
- The content lives as structured data — JSON or YAML in a repository — and the layout is a render target rather than the source of truth
- You default to a single-column layout and read the extracted text of your own PDF before sending it
- You want a diff of what changed between the version you sent in March and the one you are sending now
- You are hosting it for other people, at which point you are storing strangers' addresses, phone numbers and employment histories for no revenue and no reason
- The output has to survive an applicant tracking system you have no way to test against
- You are adding an LLM that rewrites the bullet points, because a resume that overstates what you did is a document you will be asked about in an interview
- The application deadline is tomorrow and the PDF pipeline does not exist yet
If you build it anyway
the checklist, then the prompt that enforces it
- Keep the content in a plain structured file under version control. The renderer is disposable; the evidence is not, and it should outlive whatever you build this month.
- Default to one column. Multi-column looks better to humans and worse to every text extractor, and you rarely find out which one read it first.
- Embed fonts as a subset in the PDF and verify by opening the file on a machine that has never had those fonts installed.
- Add an extraction test: pull the text back out of the generated PDF, print it, and check that dates still sit next to the jobs they belong to.
- Enforce the constraints the paid tool enforces — a minimum body size, a maximum page count, a minimum margin — and make breaking them a deliberate flag rather than a default.
- Strip PDF metadata on export, or set it to your own name. Producer and Author fields travel with the file.
- Generate a filename that is your name and the role, not resume-final-v3.pdf.
I am building a resume generator for my own use. The content is structured data
I own and the output is a PDF that will be read by parsing software before a
human sees it. Optimise for that, and push back if I ask for something pretty
that parses badly.
1. Content first. Define a schema — roles, dates, bullets, skills, education —
and store it as one YAML or JSON file. The renderer reads that file and holds
no state of its own.
2. Build the text-extraction check before the templates. After generating a PDF,
extract its text and print it in reading order, so I can see what a parser
sees. Fail loudly if the extraction is empty or scrambled.
3. Default to a single-column layout. If I ask for two columns, warn me that
extraction order goes wrong and offer to generate both variants.
4. Never rasterise text. No screenshots, no text-as-SVG-paths, no images of
headings. Every word must be selectable in the final PDF.
5. Embed and subset fonts. Tell me how to verify the embed, and use a fallback
stack that survives a machine without my fonts.
6. Enforce constraints as validation errors, not suggestions: minimum body size,
maximum page count, minimum margins. I should have to override them on
purpose.
7. Handle pagination deliberately — no orphaned headings, no role split across a
page break, and a diagnostic that tells me which entry caused the overflow.
8. Strip or set PDF metadata on export. Author is my name; Producer says nothing
about the template I copied.
9. No accounts, no server, no upload. This runs on my machine against my file.
10. Do not add an LLM that rewrites my bullet points. If I ask, remind me that I
have to defend every line of this document out loud in an interview.
11. Out of scope: cover-letter automation and bulk applying. Those are a
different problem and a worse one.That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
If the deadline is real and the pipeline is not. $21.99 is a one-off payment for a working export and, more usefully, for constraints somebody else has already argued about — the layouts genuinely do stop you making the resume worse. Building your own is the better long game because the structured file you end up with is reusable forever, but nobody has ever been hired for owning their own renderer.
$21.99/mo is cheaper than your weekend.
There is barely anything to exit. The content is a file you already own, in a format you chose, and the artefact is a PDF sitting on your disk — if the generator stops working the week before an interview, you open the last PDF, edit it in anything, and move on. Keep a rendered copy of every version you actually sent, named by date and role, so the record of what a given employer received does not depend on your code still running.
Active open-source resume builder with structured data, multiple templates and PDF export.
Renders a typeset CV from a YAML file, which is the version of this idea that survives you changing tools.
Questions
Do applicant tracking systems really fail on two-column layouts?
Some do, some do not, and you never get told which one you met. The point is that you cannot test it: the parser is on the other side of a form you only get to submit once per application. Since a single-column layout costs you nothing but a little elegance and removes the entire question, take the trade.
Is a resume builder worth building at all if the free tier exists?
The reason to build is not the templates, it is that you end up with your career as structured data. Once your roles, dates and evidence live in one file you can render a one-pager, a long-form CV and a LinkedIn summary from the same source, diff what changed since the last search, and never retype an employment date again.
What is the one thing an agent gets wrong here?
Pagination. It will build a layout that looks perfect with your current content and has no idea what happens when you add two lines — headings orphaned at the bottom of page one, a role split across the break, an accidental third page. Ask for a diagnostic that names which entry caused the overflow, and you have solved most of the maintenance in advance.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
Tracking your own applications is a table you will actually maintain because it is yours.
Comparing your CV to a job ad is text analysis on two documents you already own.
$40 a month for a board with four columns. The only sensitive thing in it is the fact that you're leaving.
last reviewed 2026-08-04 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice